What It Means That D2 Intelligence, a Wenge AI Subsidiary, Launches DocuMind for Document Verification — September 23, 2026
· CompaniesAutomation
Radar Flash Edition: D2 Intelligence, a Wenge AI subsidiary, launches DocuMind, which cross-references invoices and bills of lading and saves the proof. Tested in the Hong Kong Monetary Authority sandbox. What to do with the idea in your company.
Flash edition. D2 Intelligence, a subsidiary of China's Wenge AI, launches DocuMind: a platform that doesn't just read documents, but cross-references them and saves the proof of every check. The rest of the day in today's Radar.
What happened
The announcement came out this Wednesday, September 23, from Hong Kong (PR Newswire note, picked up by The Manila Times). DocuMind brings document recognition, data extraction, cross-verification, and risk review into a single workflow. It targets banking, insurance, and the public sector: foreign trade, credit files, KYC, underwriting, and claims. It can be installed on-premises.
The figures come from tests in the Hong Kong Monetary Authority's GenA.I. Sandbox, where D2 Intelligence is listed as a technology partner for the second cohort (27 use cases, 20 banks). A foreign trade file of up to 10 pages in less than 30 seconds, compared to 10-20 minutes manually; over 95% accuracy in key fields and around 85% automated review. Wenge AI has been listed in Hong Kong since June 26.
Why it matters
The word that counts isn't 95%, it's "cross-referenced". An OCR returns text; here, the system checks that the invoice and the bill of lading say the same thing—parties, amounts, ports, goods—and leaves a trail. That is what allows a process to be let go: a loop only runs on its own if there is an automatic verification of the result. And the figures themselves are a reminder: 95% per field means one field is wrong out of every twenty, and 15% of the review still belongs to a person. Note: these are manufacturer data in a supervisor's testing environment, not an independent audit.
For your company
You don't need DocuMind; you need the idea. In a small or medium-sized enterprise, the equivalent is the purchase order–delivery note–invoice cross-check. Three things this week:
- Choose a single high-volume workflow. Supplier invoices, delivery notes, or customer onboarding. Measure how many minutes each document costs today.
- Write the cross-check rules before buying. Which field of which document must match which: Tax ID, amount, quantities, date. These rules are your verifier; without them, any tool only extracts.
- Ask the provider for three things. Accuracy per field with your documents, where the ones that don't match go, and an exportable log of every check.
It's the jump from the integrated step to the unattended step of the AI First ladder: the machine reviews what matches, and the person only looks at what doesn't.
FAQ
What is DocuMind?
The platform from D2 Intelligence, a Wenge AI subsidiary, that extracts data from documents, verifies them against each other, and leaves an auditable trail. It is designed for banking, insurance, and administration.
Are the 95% and 85% figures reliable?
They are manufacturer figures measured in the Hong Kong Monetary Authority's sandbox, which has not published its own results. Use them as a yardstick to demand your own, not as a guarantee.
Is it useful for an SME?
The product is oriented towards Hong Kong banking. What's useful is the pattern: cross-referencing between documents, a human queue for what doesn't match, and a log of every decision.